Agricultural Systems 77 (2003) 137154

www.elsevier.com/locate/agsy

Modelling apple orchard systems

Susan M. Hester, Oscar Cacho*
School of Economics, University of New England, Armidale, NSW 2351, Australia
Received 25 May 2001; received in revised form 24 May 2002; accepted 4 July 2002

Abstract
While a number of models have been developed to assist managers of deciduous fruit tree
crops with specic aspects of decision making, most are non-optimising predictive models and
few employ detailed mechanistic models of fruit-tree growth that would enable the simulation
of any orchard system from planting to maturity. This paper details the complex biological
and economic relationships present in an apple orchard system and describes a dynamic
simulation model based on these interactions. The model is bioeconomic in nature, and may
be used to investigate a range of issues of relevance to the commercial apple orchardist. These
issues include understanding how biological factors inuence apple-tree productivity, and
how to choose among a diverse range of apple orchard systems. Each system, consisting of a
particular combination of cultivar, rootstock, tree spacing and training method, has implications for fruit quality, quantity and ultimately prot. The choice of system is made at planting, while an important annual decision is the optimal rate of thinning, both of which
determine potential yield over the lifetime of the orchard. These decisions also inuence costs
and revenues per hectare and, by necessity, are made in the context of unknown future prices
of inputs and outputs. The bioeconomic model is used to maximise net present value of one
orchard system by selecting optimal thinning strategies over a 15-year period.
# 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Horticulture; Bioeconomics; Modelling; Thinning; Apples

1. Introduction
Orchard systems are complex. Managers of deciduous perennial fruit crops must
consider both biological and economic relationships in determining preferred orchard

* Corresponding author. Tel.: +61-2-6773-3215; fax: +61-2-6773-3596.

E-mail address: ocacho@pobox.une.edu.au (O. Cacho).
0308-521X/03/$ - see front matter # 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0308-521X(02)00106-3

138

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

design and life-time orchard management strategies. Decisions about tree size and
varietal mix made at planting determine potential yield over the lifetime of the
orchard. These decisions also inuence costs and prot per hectare and, by necessity,
are made in the context of unknown future prices of inputs and outputs. Once trees
are bearing fruit, the decision concerning the amount of fruit to leave on the tree
until harvest (the thinning decision) remains one of the few ways of inuencing the
annual orchard yield.
A simulation model representing the apple orchard is a particularly useful means
of evaluating the eects on yield and protability of alternative orchard systems and
other decisions under the direct control of the orchardist. Given the importance of
biological and economic components in orchard protability, both elements should
feature in the model if the orchard system is to be adequately described and simulations meaningful. Lack of data on yields of many important apple varieties overtime means it is necessary to adopt a bioeconomic modelling approach.
Simulation models of agricultural systems have grown in popularity in recent
decades due to their usefulness in tackling the inherent dynamic and/or stochastic
nature of agricultural problems and due to increased computer capacity (Oriade and
Dillon, 1997). Simulation may substitute for large-scale physical experimentation,
which could otherwise take decades, especially in the case of perennial crops.
The biological simulation models that underlie bioeconomic models may vary considerably in their complexity and depend critically on the purpose for which they are
constructed. The biological simulation models developed by biologists and agricultural scientists are often too detailed to be used in optimising economic models
designed by economists. Conversely, simulation models developed by economists tend
to be too simple by biologists standards with models consisting of linear or simple
non-linear functions to allow for solution of optimisation problems (Cacho, 1997).
In this paper, an attempt was made to nd a balance between the complexity
demanded by biologists and the simplicity required in dynamic optimisation problems. The paper starts by presenting a review of past modelling work in perennial
fruit trees and identifying further research needs. A brief description of fruit tree
management is then presented, followed by the development of a bioeconomic
model for an apple orchard. The model is used to understand the inuence of thinning decisions on apple yields and prices and, consequently, on the net present value
(NPV) of the operation. The model is then incorporated into a dynamic optimisation algorithm to determine the optimal thinning path over a period of 15 years,
starting from orchard establishment. The paper ends with concluding comments and
future research needs.

2. Past work on modelling of deciduous fruit trees

Despite the development of a large number of bioeconomic models for use in
management of annual crops (see Penning de Vries and van Laar, 1982; Oriade and
Dillon, 1997), models are not available for perennial tree-crop research of the type
described in this paper. On the one hand, a number of biophysical models and

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

139

biological models that describe particular aspects of individual apple-tree growth

and fruit production have been developed, and only a few have been used in conjunction with an economic modelling framework. On the other hand, modelling of
perennial crops undertaken by economists include yield functions estimated from
experimental data or survey information, which often do not capture the dynamic
aspects of the system.
Models developed by economists to analyse issues in perennial-crop research have
used both non-optimising and optimising frameworks. Generally, maximising the
net present value (NPV) of the orchard, subject to aspects of orchard operation,
including optimal replacement of trees and optimal mix of tree varieties, are typical
objectives of these models. Non-optimising models tend to assess prot outcomes
resulting from a narrowly dened set of alternative management strategies.
Biologists and horticulturists have also used optimising and non-optimising models
of perennial-tree orchards. These models typically use simple economic models
attached to well developed biophysical models to analyse specic aspects of fruit production and orchard management, normally for a single growing season. Aspects of
management include alternative thinning strategies and their eects on orchard returns.
Table 1 presents a chronological review of models developed for analysis of
deciduous perennial tree crops. The list is not exhaustive; rather, it represents trends
in deciduous fruit tree research by both horticulturists and economists.
In general, it appears that the apple is a popular deciduous fruit tree for modelling
purposes. Researchers have employed simple empirical biophysical models to
describe fruit orchards and specic aspects of fruit production, and while mechanistic models of apple trees have been developed (Winters, 1976, 1986; Haley et al.,
1990), detailed bioeconomic models of orchard management are largely missing
from the research. A large number of simulation models are developed as decision
support tools for growers (Haley et al., 1990; Thiele and Zhang, 1992; Groot, 1996;
Cahn et al., 1997). While simulation features prominently as the chosen analysis
technique, dynamic programming (Childs et al., 1983), mathematical programming
(Willis and Hanlon, 1976; Graham et al., 1977) and cost-benet analysis techniques
are also used (Whitaker and Middleton, 1999). Orchard replacement policies and
fruit variety mix are the most common decision variables. Other decision variables
include planting density, fruit load, capital borrowings, rootstock, pest number and
use of hail netting. Prot maximisation and/or prot simulation is the most common
objective of the modelling work.
The models listed in Table 1 vary in their capacity to answer particular questions.
Most are non-optimising predictive models and few employ detailed mechanistic
models of fruit-tree growth that would enable the simulation of any orchard system
from planting to maturity, hence the need for the type of model described in this paper.

3. Orchard management, choices and practices

In the planting and management of an apple orchard a large number of choices
and decisions concerning the most appropriate system face the grower. Barritt

140

Table 1
A chronological sample of published tree crop models
Tree type

Biological model

Optimisation/analysis
techniques

Decision variables

Objectives

Willis and Hanlon (1976)

Apple

Childs et al. (1983)

Apple,
pear
Apple,
cherry,
pear
Apple

Dynamic linear
programming
Simulation

Vriety, capital borrowing

Winter (1976, 1986)

Survey-generated
yield information
Mechanistic,
(FRUPRO)
Survey-generated
yield information

Haley et al. (1990)

Apple

Empirical yield
estimation
Mechanistic

Johnson and Rasmussen (1990)

Thiele and Zhang (1992)

Peach
Apple

Empirical
Empirical

Groot (1996)

Empirical

Cahn et al. (1997)

Apple,
pear
Apple

Smulation/dynamic
programming
Inference procedures,
simulation
Simulation
Dnamic simulation
model
Simulation

Optimal tree mix, maximise

PV net returns
Orchard system protability
comparisons
Protability, optimise labour,
cash borrowing requirements,
renewal strategy
Maximise prot

Empirical

Smulation

Whitaker and Middleton (1999)

Apple

Yield estimation

Cost-benet analysis,
simulation

Graham et al. (1977)

Single/multiple period
linear programming

Planting density, variety,

rootstock, orchard size
Orchard system, variety and
type, timing of replacement
Replacement policy
Pest numbers
Fruit load
Fruit load
Crop protection system
Planting density, year of
rst harvest
Orchard system, erection
of hail net

Assess benets of pest control,

decision support
Maximise prot
Explore revenue per hectare,
decision support
Explore cost of pesticide
reduction, decision support
Explore NPV, IRR, decision
support
Protability of hail netting

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

Author(s)

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

141

(1987) denes an orchard system as the integration of all the horticultural factors
involved in establishing and maintaining a planting of fruit trees. Many decisions
must be made before planting occurs, including choice of cultivar (variety), rootstock (tree size), tree density and pruning and training (tree form).
The apple cultivar determines fruit variety and fruit features such as size, shape,
colour, avour, rmness, ripening season and pest and disease resistance. The age at
which an apple tree rst bears fruit (its precocity) will vary according to the cultivar/
rootstock combination.
A range of apple cultivars are normally grown in commercial apple orchards. A
variety of reasons exist for choosing a particular cultivar: each type of apple may
face distinct market situations and may potentially receive dierent prices; sequential harvest dates allow for more ecient picking and packing operations; some
cultivars have a greater susceptibility to pests and diseases than others; and dierent
cultivars may be necessary for pollination.
Tree size is central to an orchard system because of its economic implications.
Trees that are large at maturity result in increased labour costs and often take many
years to bear fruit. An objective of modern orchard design is to maintain smaller
trees that are planted closer together. These smaller trees have the additional
advantage of earlier bearing habits. Tree size is most often controlled physiologically
through rootstock selection.
Horticultural techniques used to manage vegetative growth and fruit production
are pruning, training and thinning. While the level of thinning is determined each
year, pruning and training are usually predetermined by choice of density, rootstock
and cultivar when the orchard is established. These latter choices are made at
planting and assumed to remain unchanged during the life of the orchard.

4. The model
4.1. Conceptual biophysical model
In this paper, apple tree growth during a growing season and over a lifetime is
described using a carbon-balance model. Seasonal growth patterns that are peculiar
to deciduous woody perennials, such as apple trees, form an integral part of the
model. The carbon-balance model is described in detail in Hester (2000) and only a
brief overview is given here.
Fig. 1 shows the conceptual biophysical model. The critical environmental variables in the model are day length, temperature and lightall three aect the growth
potential of the tree. Light (radiation) and daylength aect canopy photosynthesis,
while temperature aects both photosynthesis and respiration. Light interception is
critical to the amount of photosynthesis undertaken by an apple tree, and the level
of interception is determined by shape of the tree which, in turn, is inuenced by
chosen pruning and training techniques and tree age. Net photosynthesis describes
the amount of carbon available for tree growth after accounting for carbon lost
during leaf respiration and gained through photosynthesis.

142

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

Fig. 1. Conceptual biophysical model.

Photosynthesis provides energy used for growth by each tree component: roots;
wood; fruit; and leaves. Respiration of these components results in energy losses.
The fruit load on the tree is the major determinant of how energy is divided
between each tree component and is inuenced through thinning.
Fruit load also inuences size of individual fruits and has an impact on future fruit
production through its eect on the biomass of root, wood and leaf that grows in
the current and following seasons. Fruit is harvested annually and its quantity and
quality is taken into account by the model.
Apple trees follow temperature-driven patterns of growth and non-growth over a
12-month period. During late autumn, apple trees enter a dormant period where
vegetative and reproductive buds require a period of chilling temperatures if bloom,
growth and development are to occur in the spring. Dormancy is broken once the
chilling requirement of the particular apple cultivar has been satised. Growth,
however, does not automatically resume at this point, rather, it occurs following
higher daily temperatures that result in accumulation of a certain number of heat
units. These events are accounted for in the model by a set of triggers associated
with heat and chill unit accumulation.
4.2. The bioeconomic model
The economic model describes the costs and revenues associated with fruit production in an orchard system from planting to maturity. Annual fruit production is

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

143

determined by the biophysical model and used in the economic model to simulate
the annual protability and net present value (NPV) of a particular system.
The relationship between the biophysical and economic models is described in
Fig. 2. The yield and NPV of each orchard system will depend fundamentally on the
variety, tree density and pruning/training regimes chosen when trees were planted.
In addition, yield and protability may be signicantly modied each year through
changing the amount of thinning undertaken. Thinning is used to modify the number of fruits a tree bears to maturity and thus it determines their size and inuences
the price they receive. The length of the time horizon, T, also inuences system NPV
and can be included as a decision variable. The time horizon is important in assessing orchard rotation strategies, however, is not considered as a decision variable in
the current model.
The decision variables can be viewed as being of two types. Those pertaining to
the establishment of the orchard, and those management decisions made annually.
Thinning falls within the latter category.
Annual yield is used in the economic model to determine annual revenue from the
system. Apple prices are given exogenously. The cost of establishing the orchard
system as well as costs related to tree density and harvest enter the NPV calculation.
When the model is used in optimising mode, decision variables are adjusted so that
NPV is maximised.
The main features of the model are detailed more formally in this section. The net
present value (V) of the stream of annual prots obtained over the planning horizon
t=1,. . .,T is dened as:

Fig. 2. Diagrammatic representation of the bioeconomic model (j=orchard system, t=year).

144

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

T
X

t Ft fxt ; ut g1 rt CE

t1

where the annual prot
t is a function of fruit production per hectare (Ft), r is the
discount rate and CE is the cost of establishing the orchard starting with bare
ground. Fruit production is, in turn, determined by a state vector xt and a control
variable ut representing thinning. The variable ut is inversely proportional to thinning intensity, it takes on a value of zero when thinning is 100%, so that no fruits
remain on the tree, and it takes on a value of one when no thinning occurs, so that
all fruits remain on the tree. Values of ut between zero and one indicate intermediate
levels of thinning. In the economic model, time (t) is a discrete variable measured in
years. Annual prot is described as:

t Pa fFWt g  CFFt fxt ; ut g  NCNt fAt g

where Pa is the price of fruit (A$/kg) which is a function of average fruit weight
(FWt ) and measured in grams of dry matter, CF is the cost of harvesting the fruit
(A$/kg), N is the planting density (trees/ha), and CNt represents costs per tree.
Labour is the main component of harvesting cost (CF). Cost per kilogram harvested
depends on variety and orchard system used, because small, compact trees require
less labour per kilogram harvested than large trees. Density-related costs (CNt)
depend on age of the orchard (At) and consist of labour, materials, chemicals, fertiliser and other expenses, such as irrigation costs per tree. These costs are assumed to
increase at a decreasing rate, hence:
dCN
> 0;
dAt

and

d 2 CN
40
dAt2

The choice of orchard system determines pruning and training requirements and
ultimately aects the shape of the trees which, in turn, may have a bearing on time
and eort required to apply chemicals, fertilisers and irrigation. In summary, all
costs (CE, CF and CNt) depend on the orchard system selected. The model represents a single orchard cycle starting with bare ground, thus At=t.
Annual fruit production (kg/ha) is determined by photosynthesis and respiration
rates during the year:
t1
Ft N

G  L C
F; xt ; ut d

t

where  represents time measured in days, G is daily photosynthesis, L is daily

respiration, C represents carbohydrate reserves mobilised in day , and F,, is the
proportion of dry matter allocated to fruit on a daily basis. The partitioning parameter (F, ) depends on state and control variables, as described later in this section.
The integrand of Eq. (3) represents the daily amount of carbon partitioned to fruit

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

145

production and growth. The parameter  converts carbon to fresh fruit weight. Thus
the integral represents fruit production per tree in the interval from the beginning to
the end of year t, and this is multiplied by tree density (N) to obtain yield per hectare.
The biophysical model consists of four state variables: leaf, fruit, wood and root.
Each of these variables is associated with a dierential equation for daily gain in
mass of each component. Dierential equations are numerically integrated on a
daily basis in the simulation. From the standpoint of the economic model, leaf and
fruit are not treated as state variables since they follow an annual cycle that, for a
given orchard system and climatic events, depends on the amount of carbohydrate
reserves available in wood and root at the beginning of the season. Thus, the state
variable vector for each year is dened as:
xt Rt ; Wt ; Dt

The state of an individual tree at the beginning of year t is described using dry
matter content in root (R) and wood (W) and the presence of pests and diseases (D).
The annual objective of thinning, to achieve the most protable fruit size, is tempered by the eect of current fruit load on potential fruit size and yield in future
years. If a tree bearing a large number of apples is thinned only lightly, it is probable
that the tree will only be able to support a small number of apples in the following
year, which may negatively impact on prots. This tendency towards biennial bearing is inherent in many apple cultivars and may be modied by thinning. The orchard manager must therefore take into account both immediate and long-term
fruiting prospects of the tree when determining the appropriate level of annual
thinning to maximise NPV.
Following picking, apples are graded and packed into various count sizes that
are based on individual fruit weight and reect number of fruit per carton. The value
of the apple harvest is closely related to fruit size. Generally a crop of small apples is
worth less than the same weight of larger apples, hence the incentive to improve fruit
size through thinning.
In the model, a biennial bearing pattern is imposed on fruit growth that sees a
large crop followed by a light crop. Data on natural fruit bearing patterns is scarce
although the biennial bearing pattern simulated by the biophysical model does
compare reasonably with the fruit mass data from an Australian trial (Middleton,
1984; Middleton, unpublished data). The biennial bearing pattern is simulated as
follows:
EFLt BFLFLt1  BFL

where EFLt is expected fruit load when no thinning takes place, BFL is fruit load
that would lead to minimal or no biennial bearing and FLt is fruit load in the previous year. Fruit load is dened as the number of apples per m2 of leaf area. The
value of EFLt depends on the amount by which actual fruit load exceeded BFL in
the previous year. The estimate of BFL is set at a level that appears reasonable given
data from an Australian trial (Middleton, 1984; Middleton, unpublished data).

146

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

The process of thinning leaves a proportion of the original amount of fruit on the
tree. The actual fruit load that results from thinning (FLt), is calculated as:
FLt EFLt ut 0 4 ut 4 1

The energy for growth produced as a result of photosynthesis, in the form of dry
matter, is allocated to fruit, leaves, wood and root components of an apple tree. This
allocation process is known as partitioning, and is aected by whether or not a tree
is producing fruit. One noticeable eect of the presence of fruit is a reduction in dry
matter partitioned to root and hence a reduction in tree growth (Heim et al., 1979).
A heavy fruit load (FL), dened as fruit number per m2 leaf area, reduces photosynthesis through a reduction in leaf area (shoot growth) when compared with trees
having no fruit (Faust, 1989).
Heim et al. (1979) found that levels of partitioning of dry matter to various tree
components were caused almost entirely by dierences in fruit load. Using data
from Heim et al. (1979), the proportion of assimilates partitioned to tree components during a growing season was estimated in the model using the following
equations:
j j

lj FLt
j FLt

R 1  L  W  F

for j L; W; F

7
8

where j is the proportion of dry matter partitioned to the various tree components
and lj ; j and j are constants.
The pattern of partitioning is important in perennial tree crops such as apples,
because the relative dry matter allocations to tree parts in the current year have
implications for growth and fruiting potential in following years. At the extreme, a
tree may not bear a regular crop every year but instead a heavy crop may be followed by a small crop because of fruit overload in a particular year (Monselise and
Goldschmidt, 1982). Inuencing the partitioning process in an apple tree through
the adjustment of fruit quantity may correct these biennial bearing patterns if a
cultivar is prone to them. One important outcome that normally results from lowering fruit load of a tree is improvement in average fruit size.
Once FLt is determined, fruit number (FN) is calculated according to:
FNt FLt : LAmaxt

where LAmaxt is the maximum leaf area (m2 leaf area), calculated from the biophysical
model. Thus the units for FN will be a number, representing the total number of
apples on a particular apple tree. Finally, average fruit weight (FWt) is calculated as:
FWt

YF t
FNt

10

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

147

where YFt is the yield per tree measured in grams of fresh weight, calculated in the
biophysical model. Average fruit weight is used in the economic model to determine
the protability of a given thinning strategy.

5. Model implementation
In this section, the protability of one hectare of Granny Smith apple trees is
simulated over a planning horizon of 15 years from planting. The trees are planted
on the semi-dwarng MM.106 rootstock, at a density of 1000 trees per hectare and
trained to the central leader system. The specic system chosen for simulation is one
of those outlined in Hester (2000).
Prices received by the apple orchardist are determined exogenously. For a given
variety, apple prices vary according to individual fruit size. Apple size is approximated by the fruit count or number of apples in an 18-kg carton. The higher the
count the smaller the apple size (weight) and vice versa. Larger sizes receive a price
premium; however, enormous apples receive a price penalty.
Data on prices received for each grade of apple variety are not regularly recorded
at fruit selling centres in Australia, rather, only maximum, minimum and average
prices for a variety are publicly available. Detailed price by grade data was obtained
for Granny Smith apples, sold during January 1999 and was used to estimate the
grade price dierentials associated with fruit size for all varieties.
For simplicity, it is assumed that the relative price dierences shown in Table 2
hold regardless of maximum price. The data were used to estimate the following
price-count relationships used in the economic model:
8
P75
if countt 4 75
<
Pt
Pmax
if 75< countt <105
11
:
c  0:2448countt if countt 5 105

Table 2
Price data used to develop price-count relationship
Count

Price per 18 kg carton ($)

70
80
90
100
110
120
135
150
165
180
198

36
43
43
43
38
31
25
24
21
16
15

148

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

where P75 is the price for counts less than 75, Pmax is the price received for counts
between 75 and 105 and c is an intercept term. Using the Batlow data P75 has a
value of A$36 per carton, Pmax is A$43 per carton and c is 61.29.
Costs related to tree density (CN) are presented in Fig. 3 and are specic to the
chosen system. Labour costs peak in year four after planting when pruning and
training requirements are highest. After this time, labour requirements remain
stable. Chemical and nutrient costs are lowest during the years immediately after
planting but increase as the tree reaches maturity. The peaks in nutrient costs
represent the cycle of applications that is adopted in this orchard. Capital costs for
training are minimal in the central leader system and are incurred in the 2 years
immediately after planting.
Establishment costs are estimated to be A$8770 per hectare for this orchard system, assuming a cost of A$6.25 per tree purchased from a commercial nursery.
The biophysical model, described by Eq. (3) is solved using numerical integration
at daily steps and is implemented using the specialised simulation packages Simulink1 and Matlab1 (Mathworks, 1996, 1997). The biophysical model is solved for
specic values of thinning and the resulting fruit quantities and sizes are used in the
economic model to calculate annual prot and NPV.

Fig. 3. Density-related costs (CN) over the simulation period.

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

149

6. Results and discussion

6.1. The eect of thinning
Before undertaking dynamic optimisation, the eect of thinning was investigated
in more detail. Simulated trees were thinned so that various amounts of fruit
remained on the tree, and the eects of a given thinning rate, held constant over 15
years, were investigated. As thinning rates were varied for each 15-year cycle, their
eect on fruit weight, yield, price and per hectare NPV were analysed. Results of this
analysis are shown in Fig. 4.
Average fruit weight decreases in a non-linear fashion (Fig. 4a) as a greater number of fruit is retained on the tree (thinning level decreases). This relationship reects
the competition between individual fruits for the given amount of energy available
for fruit growth. Although available energy for fruit growth increases as the fruit
load on the tree increases, this additional energy does not compensate for the
increased demands of a greater number of fruit, hence the decrease in average fruit
weight.

Fig. 4. The eect of thinning target (u) on average fruit weight (a), price (b), yield (c) and NPV (d).

150

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

The decrease in average fruit weight has implications for the average price
received (Fig. 4b). When fruit is thinned so that only a small number of fruits remain
on the tree (u=0.05) average price is slightly above A$2.00 per kilogram. At this
level of thinning average fruit weight, and hence size, is greatest. Highest prices are
not received here because the largest sized apples (count > 75) do not receive the
highest price per kilogram. Fruit price does increase as more fruit are left on the tree
and a drop in average weight (size) improves their value. Apples receive the highest
price per kilogram when thinning leaves between 20 and 25% of fruit on the tree
(0.24u 40.25). Beyond this level of thinning average fruit price decreases in line
with reductions in fruit size that put the fruit in lower-valued count sizes.
While leaving progressively more fruit on the tree does reduce average fruit
weight, it also has the eect of increasing total fruit weight on the tree, or yield
(Fig. 4c). As thinning increases, yield per tree also increases, but at a decreasing rate.
When thinning leaves only around ve per cent of fruit on each tree (u=0.05), total
yield is approximately 20 kg per tree, and increases to around 75 kg of fruit per tree
when thinning leaves around 55% of fruit on the tree (u =0.55). However, at high
yields, average fruit size is small and average price per kilogram of fruit is well below
the maximum.
Ultimately, the value of thinning hinges on its eect on NPV. As u increases from
0.05 to 0.45, NPV increases (Fig. 4d) to a maximum of A$1,210,102 and decreases
thereafter. Prior to the level of thinning that produces maximum NPV (u=0.45), the
increased yield per tree compensates for the reduced price per kilogram received for
the progressively smaller fruit. However, beyond u=0.45 average fruit size is
reduced to a level that receives dramatically lower prices, and the increase in average
yields per tree does not compensate for this price drop.
It is interesting to note, therefore, that maximum NPV does not occur where
prices are at a maximum. Nor does it occur where fruit yield or fruit weight are at
a maximum. Rather, the trade o between total yield per tree and individual fruit
weight results in the achievement of maximum NPV where prices are below the
maximum level.
This simple analysis provides insight into the operation of the model and the eect
of thinning. However, these results only apply when thinning is treated as a static
variable, taking the same value every year.
6.2. Dynamic optimisation
Initial attempts at solving the dynamic model with a nonlinear programming
(NLP) algorithm, showed that the solution was highly sensitive to the starting point
for the search. This suggests that the model tends to converge to a local optimum,
this problem is described by Cacho (1998). To overcome this problem, a genetic
algorithm (GA) was combined with the NLP in an attempt to identify the global
maximum. The GA was adapted from that reported by Cacho and Simmons (1999).
The GA was rst used to undertake a global search of the decision space and, once
the GA converged, the NLP was activated using the GA solution as the starting
point. This approach yielded a higher NPV than any solution of the NLP on its

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

151

own. Although there is no guarantee that the global optimum was found, the technique increased our condence that the solution was close enough to the best possible for practical purposes.
The optimal thinning trajectory is shown in Fig. 5. This trajectory gives a cumulative fruit yield of 967 kg/tree over the 15-year period with an NPV of A$1,265,000/
ha over the same time horizon.
The optimal trajectories of fruit weight, price, yield and prot are presented in
Fig. 6. Average fruit weight remains constant after year four (Fig. 6a) and maximum
prices per kg are received in all but the second year from planting (Fig. 6b). Average
fruit yield per tree (Fig. 6c) rises consistently over time, although the increases in
years 24 are slower reecting the thinning regime during this time, when relatively
large amounts of fruit were left on the tree. Annual discounted prots (Fig. 6d) are
aected by fruit prices resulting from the optimal thinning strategy. Before the trees
begin to bear fruit, initial establishment costs and variable costs result in a negative
prot. Prot continues to rise until year 10 when it levels out and remains reasonably constant for the remainder of the planning horizon.
Optimal fruit yield and fruit load results over 15 years are compared with a
situation of no thinning in Fig. 7. When no thinning takes place, the biennial bearing pattern that results from the model shows a very clear and consistent pattern of
alternation between the o year and the on year of fruit production. In the on
years, yield increases from 18 kg per tree in year two to around 200 kg per tree in
year 14. In the o years, no fruit is produced due to the eect on the tree of the
high yields in the previous year (Fig. 7a). This contrasts with fruit production per

Fig. 5. Optimal thinning targets (u*) using an NLP (dotted line) and the GA-NLP algorithm (solid line).

152

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

Fig. 6. Optimal trajectories for fruit weight (a), fruit price (b), fruit yield (c) and discounted prot (d)
under an NLP (dotted lines) and the GA-NLP algorithm (solid lines).

Fig. 7. Optimal trajectories (solid lines) for fruit yield (a) and fruit load (b) compared with the no-thinning option (dotted line).

tree using the optimal thinning strategy, where the cycle of dramatic uctuation is
eliminated. When no thinning occurs, the on years produce a fruit load of 20 fruits
per m2 of leaf area, and a fruit load of zero in the following year (Fig. 7b). The
optimal thinning strategy modies these dramatic uctuations and fruit load
appears to settle at a value of six fruits per m2 of leaf area. Small uctuations occur
in the earlier years but are reduced to almost no variation by year ten.

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

153

7. Summary and conclusions

The lack of a suitable model to analyse a range of production issues confronting
an apple orchard manager led to the development of a bioeconomic model discussed
in this paper. The model allows simulation of the eect of management decisions on
any apple orchard system from planting to maturity. The analysis in this paper
focuses on optimal thinning rates over the lifetime of the orchard given an inherent
tendency towards biennial bearing of many apple tree cultivars.
Thinning was rst investigated through simple simulation experiments, where its
value remained static in each year of the 15-year planning horizon. As thinning rates
were varied for each 15-year cycle, its eect on fruit weight, yield, price and per
hectare NPV were analysed. It is interesting to note that maximum NPV did not
occur where prices were at a maximum. Nor did it occur where fruit yield or fruit
weight were at a maximum. Rather, the strong trade o between total yield per tree
and individual fruit weight, which results in better prices, resulted in the achievement of maximum NPV where prices were below the maximum level.
The model was then incorporated into a dynamic optimisation algorithm consisting of combining a genetic algorithm (GA) and a nonlinear programming (NLP)
model. The objective function was maximisation of NPV by setting annual thinning
rates. After initially uctuating, presumably to null the eect of the tendency
towards biennial bearing, the optimal level of thinning remained reasonably constant from year 5 to 15. Yield per tree gradually increased as did the discounted
value of prot. In all but one year, the optimal thinning strategy resulted in fruit
growing to a size that received the maximum price per kilogram. The NPV obtained
under dynamic optimisation was 5% higher than that obtained under static optimisation (where thinning rates remained constant throughout the planning horizon).
Incorporating the biennial bearing habit into the model and optimising decisions
that inuence it, represents a signicant contribution to understanding the apple
orchard problem. The model can be used to compare the economic performance of
dierent orchard systems under optimal management, which is important in the
initial planning stage before orchard establishment. Variables of interest include apple
variety mix, tree density, pruning and training system. An important extension for
future research is the incorporation of stochastic elements by allowing solar radiation, temperature and prices to change according to known probability
distributions.

References
Barritt, B.H., 1987. Orchard systems research with deciduous trees: a brief introduction. HortScience 22
(4), 548549.
Cacho, O.J., 1997. Systems modelling and bioeconomic modelling in aquaculture. Aquaculture Economics and Management 1 (1), 4564.
Cacho, O.J., 1998. Solving bioeconomic optimal control models numerically. In Gooday, J. (Ed.),
Proceedings of the Bioeconomic Workshop. Workshop (Post Australian Agricultural and Resource
Economic Society Conference), 22 January, Armidale, NSW, ABARE, Canberra.

154

S.M. Hester, O. Cacho / Agricultural Systems 77 (2003) 137154

Cacho, O., Simmons, P., 1999. A genetic algorithm approach to farm investment. Australian Journal of
Agricultural and Resource Economics 43 (3), 305322.
Cahn, M.B., Stevens, R.B., van de Klundert, A.C.A., 1997. Economic outcomes of early cropping and
tree decisions in Braeburn apples on Malling 9 rootstock. Acta Horticulturae 451, 551558.
Childs, R.A., Milligan, R.A., White, G.B., Stiles, W.C., 1983. A dynamic programming approach to apple
orchard replacement. A. E. Res. 8311. Department of Agricultural Economics, Cornell University,
Ithaca, 74p.
Faust, M., 1989. Physiology of Temperate-Zone Fruit Trees. Wiley-Interscience, New York.
Graham, J.D., Kennedy, G., Marshall, A.A., 1977. A Multi-period Orchard Management Model. Economics Branch, Agriculture, Canada, Vancouver.
Groot, M.J., 1996. A decision-support system for fruit crops for economic and ecological calculations.
Acta Horticulturae 249, 359365.
Haley, S., Currans, R.G., Croft, B.A., 1990. A computer aid for decision making in apple pest management. Acta Horticulturae 276, 2227.
Heim, G., Landsberg, J.J., Watson, R.L., Brain, P., 1979. Eco-physiology of apple trees: dry matter production and partitioning by young Golden Delicious trees in France and England. Journal of Applied
Ecology 16, 179194.
Hester, S.M., 2000. Evaluating Apple Orchard Management using a Bioeconomic Model. PhD thesis,
University of New England, Armidale, Australia.
Johnson, R., Rasmussen, J.M., 1990. Peach thinning optimisation model. Acta Horticulturae 276, 247
255.
Mathworks, 1996. Using Matlab. The Mathworks Inc, Natick, Massachusetts.
Mathworks, 1997. Using Simulink. The Mathworks Inc, Natick, Massachusetts.
Middleton, S.G., 1984. Pome fruit rootstock selectionapple virus free close plant Trial A. Final Report.
Granite Belt Horticultural Research Station Annual Report, Stanthorpe.
Monselise, S.P., Goldschmidt, E.E., 1982. Alternate bearing in fruit trees. Horticultural Reviews 4, 128
173.
Oriade, C.A., Dillon, C.R., 1997. Developments in biophysical and bioeconomic simulation of agricultural systems: a review. Agricultural Economics 17, 4548.
Penning de Vries, F.W.T., van Laar, H.H. (Eds.), 1982. Simulation of Plant Growth and Crop Production. Simulation Monographs. Centre for Agricultural publishing and Documentation, Wageningen.
Thiele, G.F., Zhang, J., 1992. The dynamic apple tree system: relationships to aid management strategies.
Acta Horticulturae 313, 249255.
Whitaker, K., Middleton, S., 1999. The protability of hail netting in apple orchards. paper presented at
the 43rd Annual Conference of the Australian Agricultural and Resource Economics Society,
Christchurch, 2022 January.
Willis, C., Hanlon, W., 1976. Temporal model for long-run orchard decisions. Canadian Journal of
Agricultural Economics 24, 1728.
Winter, F., 1976. A simulation model for studying the eciency of apple and pear orchards. Gartenbauwissenschaft 41, 2634.
Winter, F., 1986. FRUPRO: A bio-economic simulation model for apple orchards. Acta Horticulturae
184, 199208.